Vision-based detection system for recovering recyclable materials from dust-affected engineering waste streams

Rapid urbanization has generated large volumes of engineering waste, creating an urgent need for efficient recovery of recyclable materials to support sustainable construction and recycled material production. In practical sorting facilities, engineering waste streams are often cluttered, densely stacked, and affected by surface dust, which obscures visual features and reduces the reliability of automated identification. To address these challenges, this study proposes a vision-based detection system for recycling-oriented sorting of dust-affected engineering waste on an intelligent sorting platform. A tailored detection framework is developed to enhance feature robustness under dust interference, dense stacking, and complex background conditions. The proposed system achieves reliable real-time performance, reaching 98.82% mAP at 44.1 FPS, with notable improvements over the baseline model. The system has been deployed on a robotic sorting platform, enabling accurate identification and classification of recyclable materials to support automated, resource-efficient waste sorting for sustainable construction within the built environment.